What problem does it solve? Long, multi-phase tasks often lose alignment as context grows: goals drift, assumptions go unrecorded, and progress becomes unrecoverable across sessions. This Skill provides a heavyweight task management system that anchors intent, tracks checkpoints, and enforces acceptance criteria for genuinely complex work. ## Core Features & Use Cases - Grounding Anchor (anchor.md): Captures intent, critical constraints, scope, Done-when criteria, and assumptions with versioning, serving as the single source of truth against hallucination and drift. - Event-Driven Checkpoints with Re-grounding: Updates checkpoint.md on key events (file changes, subtask completion, new instructions) with item-by-item verification against the anchor, plus rolling compression of old records. - Lifecycle Scripts: Python scripts (task-lifecycle.py, session-catchup.py) create, suspend, resume, validate, archive, and summarize tasks in a .taskflow/ directory with schema validation and atomic writes. - 3-Strike DEADBAND Protocol: After three failures on the same problem, resets strategy to a materially different recovery route instead of silently retrying or prematurely escalating. - Use Case: When refactoring a 2000-line legacy module across multiple sessions, use this Skill to record the intent and constraints once, checkpoint progress after each extracted submodule, and resume cleanly in a new session without re-explaining the task. ## Quick Start Ask the agent to use unified-taskflow to manage your complex multi-phase task, and it will create an anchor and checkpoint files under .taskflow/ to track the work.
